Chaotic characteristic analysis for traffic flow series and DFPSOVF prediction model

Abstract

Time delay and optimal embedding dimension for the real measurement traffic flow series, which are used by mutual information method and false nearest-neighbor method, respectively, are determined for phase space reconstruction of the traffic flow series. The saturation correlation dimension and the largest Lyapunov exponent for traffic flow series are calculated to estimate its chaotic characteristics. Based on the least mean square (LMS) algorithm, a novel second-order Volterra model using Davidon-Fletcher-Powell method (DFPSOVF) is constructed, in which a variable convergence factor based on a posteriori error assumption, characteristic of real-time change with the input signal, is applied. DFPSOVF model can avoid some problems caused by improper selection of parameters when using LMS adaptive algorithm for coefficient adjustment in Volterra model. DFPSOVF model can also be applied to short-term traffic flow prediction with chaotic characteristics. Results show that when model memory length is consistent with embedding dimension of traffic flow series, it obtains higher prediction accuracy, which can meet the needs for traffic guidance and traffic control, and can also provide a new method, a new idea and engineering application reference for intelligent traffic control.